Distributed communication systems or networks are now widely used. Such communications systems may be widely distributed information sources such as local- or wide-area networks, corporate or military communication systems, or the world-wide computerized interactive communication system known as the internet. Such communication systems can be used to access remote information or computing resources. When using such systems, it is necessary to know the location or address of the information being sought or of the computing resource to be used, or the user must carry out a search procedure to locate the information. In the case of very large distributed systems which are not indexed, the search is so extensive or time-consuming that it may be impractical, and may therefore not be attempted.
There has been interest in the use of intelligent mobile agents for overcoming some of the problems associated with searches of distributed, possibly unindexed, data bases or computing resources. An intelligent mobile agent is a computer program which can independently or semi-independently perform tasks which the operator could not perform on his own because of the time or effort required for the task. For example, an intelligent mobile agent might be used to discover information autonomously, or under its own control, because the operator or user of the distributed communication system might not be aware of the existence of a source of useful information, or of its address if its existence were known. The mobile intelligent agent is a computer-type program characterized by the ability to move over the communication system from one computer to another while the program is "running"; in fact, it is necessary for it to do so in order to perform its tasks. This does not, of course, mean that the bit pattern of the program changes while it is in transit from one computer to the other, but rather means that, in performing its tasks or "running," the program moves its location from one computer to another. Intelligent mobile agents or "softbots" are described in the following references:
1) "Intelligent Agents: The First Harvest of Softbots Looks Promising", Sara Reese Hedberg, IEEE Expert. PA0 2) "Harnessing the World Wide Web", Jay Allen Sears, US Advanced Research Projects Agency, IEEE Expert. PA0 3) "Intelligent Agents on the Internet: Fact, Fiction, and Forecast", Oren Etzioni and Daniel S. Weld, University of Washington, IEEE Expert. PA0 4) "AI on the WWW--Supply and Demand Agents", Carol Brown, Oregon State University; Les Gasser and Daniel E. O'Leary, University of Southern California; Alan Sangster, University of Aberdeen, IEEE Expert. PA0 5) "Telescript Technology: Mobile Agents", James E. White, General Magic, Inc., American Association for Artificial Intelligence.
One problem which has been found with the use of such mobile agents for seeking information on a distributed computing or knowledge base is that the capability of the agent is more or less related to the size or complexity of the program which embodies the intelligent mobile agent, so that a highly capable mobile agent, just like a highly capable computer program of any sort, tends to be very large in terms of the number of bytes which it contains. Thus, autonomous agents which maintain their state over time, and which are not totally "scripted" and therefore are knowledge-based or "learning" programs, as are the domains of knowledge about which they maintain an understanding, tend to be quite large. The length of time required to transmit the intelligent mobile agent from one computer to another (from one "site" or network "node" to another) over the communication system depends upon the bandwidth of the system in bits or bytes per second. It may not be practical to use a highly capable intelligent mobile agent, because the large size of such an agent requires excessive computer-to-computer transmission time. If the bandwidth of the distributed communication system can be controlled, then its bandwidth can simply be expanded in order to accommodate the desired mobile agents at the desired speed. Ordinarily, the communication system or data base is a given, and its bandwidth cannot be controlled. It should be noted here that limitations in the bandwidth of a part of a computer which interacts at a network node with the communication system will have the same effect, as to that computer, as a limitation of the bandwidth of the communication system itself; it is therefore desirable to use the highest possible speed modem at each computer site. Under bandwidth-limited conditions, the use of a highly capable intelligent mobile agent may result in excessive transmission time delays, and these directly impact the time required for the intelligent mobile agent to perform a search. The amount of time which is considered to be excessive may depend upon the information sought and its importance. If search time is irrelevant, a highly capable intelligent mobile agent may be used on a low-bandwidth distributed computing system or network. Ordinarily, however, large transmission times are undesirable because of the resulting delays in receiving the search results, especially in those cases in which real-time information is sought. Thus, if a very complex search is required to find the desired information, a highly capable intelligent mobile agent capable of performing the search may be too large to be practical under given circumstances. The inability to quickly transmit highly capable intelligent mobile agents tends to prevent full utilization of the resources of the communication system.
U.S. patent application Ser. No. 08/741,759, filed Nov. 5, 1996 in the name of Whitebread et al., describes a scheme for improving the resource utilization in limited bandwidth communication systems. As described generally therein, a distributed communication network includes a plurality of computers, and communication paths for coupling the computers together at nodes of the network for communications therebetween. The network also includes standardized computer network software associated with each of the computers, for providing basic network communications between or among the computers; this may be software suitable for use with a world-wide communication system commonly known as "internet." The network according as described in the Whitebread et al. application also includes a mobile agent generating program located at one or more of the computers of the network, for generating, and for transmitting over the communication network to at least one other computer of the network, agents for performing a desired function. The dimensions (size) of a mobile agent are related to its capability, whereby highly capable agents may be so large as not to be usable, because of network bandwidth limitations. A mobile agent docking arrangement is located at each computer or node of the network, which capability expander is capable of coacting with the mobile agents, for rendering the agents active at that computer. A mobile agent capability expander is located at, or associated with, at least one of the computers of the network which is capable of coacting with mobile agents, for extending at least one capability of a mobile agent. This enhances the capability of mobile agents used in the system without expanding limited network bandwidth. The information embodying the capability enhancement or expander may be transmitted over the communication network to the agent-receiving computer(s) at infrequent intervals, during which the communication network is otherwise lightly loaded, so that the capability expander is available at such later time at which a mobile agent arrives. The capability enhancement may instead be uploaded to the agent-receiving computer(s) from a portable memory, such as a floppy disk.
FIG. 1 illustrates a communication system designated generally as 10, which includes a network 12 and four representative network nodes with computers 14a, 14b, 14c, and 14d connected to the network, as described in the abovementioned Whitebread patent application. As mentioned above, each computer is capable of sending intelligent mobile agents out to any of the other three computers. Such mobile agents can be generated at each of computers 14 by the use of a language such as agent TCL, available from the computer science division of Dartmouth College, Dartmouth, N.H., and the mobile agent can be transmitted, by way of the network 12, to the other computers after its generation. The bandwidth of a network such as network 12 varies from about 4800 baud or bits-per-second to as much as tens of megabaud, and may vary within the network, depending upon which network branch is considered.
FIG. 2 is a simplified block diagram of an arrangement which is similar to a portion of the arrangement of FIG. 1. In FIG. 2, a source of intelligent mobile agents is illustrated as 14a, which is the same designation as that of computer 14a of FIG. 1. Element 14a of FIG. 2 is designated as a "source", because in the most basic system, it needs only to transmit or launch a mobile agent; it may, of course, have other capabilities. The active agent goes out onto the network from element 14a of FIG. 2, and performs its tasks, moving from computer to computer as necessary. If the mobile agent is programmed to return to source 14a when its task is completed, then source 14a must also be capable of receiving the information conveyed by the mobile agent. While source/sink 14a of FIG. 2 need not be a computer, it will ordinarily be such.
Also in FIG. 2, network 12 is connected to a second "computer" block 14b, which may include a database 210 and an agent dock 212. Database 210 may be simply a memory which can be accessed by an intelligent mobile agent, in which case block 14b must further include a processor or "computer", because the intelligent mobile agent must reside, at least temporarily, at location 14b. Instead of a database such as 210, agent dock 212 could be coupled to any other data or computing resource which is to be acted upon by an intelligent mobile agent. As also illustrated in FIG. 2, database 210 is coupled to agent dock 212, and the network 12 is also connected to agent dock 212.
The intelligent mobile agent is a program. In an ordinary computer system, blocks 210 and 212 of location 14b of FIG. 2 may be hardware, or they may be programs (software) which are resident at location 14b. Agent dock 212 of FIG. 2 has the ability to interface or interact with the intelligent mobile agent, and also provides the intelligent mobile agent with capabilities which reside at location 14b. Such capabilities might include access to knowledge bases, additional computational functions, and the like.
FIG. 3 is a simplified architecture diagram which represents the software in the agent dock 212 of FIG. 2. In FIG. 3, one or more intelligent mobile agents 310a, . . . , 310n appears at a layer 310 at the bottom of block 212, adjacent to the path 214 by which it (they) arrive at the dock. It should be recognized that at any particular time, the agent dock may have no mobile intelligent agents docked therewith, it may have only one, or it may have two or more. When docked, each mobile agent 310a, . . . , 310n interfaces with a docking software layer 312, which recognizes the presence of the mobile agent, and which enables the mobile agent by effectively "pushing its ON button", to cause the intelligent mobile agent to recognize that it has arrived at a new location, and to enable it to execute its program or instructions. Location 14b also includes a knowledge base illustrated as 314.
Following the turn-on or enabling of the then-resident intelligent mobile agent 310a, . . . , 310n of FIG. 3, the docking layer 312 acts as a service provider to the intelligent mobile agent 310a, . . . , 310n. For example, the (or one of the) intelligent mobile agent(s) may indicate to the docking layer that the mobile agent is a search agent which is looking for particular data. Such data, in a military context, might be the presence, location, or identity, or all of this information, relating to electromagnetic radiators which might have been located by specialized equipment available at location 14b. It may happen that the computer at location 14b has no specialized equipment for generating the type of data which the intelligent mobile agent seeks, in which case the docking layer 312 of FIG. 3 so advises the agent 310a, . . . , 310n. In that event, the agent may go elsewhere in the network (to another computer location 14a, 14b, 14c, 14d, . . . ; of FIG. 1), or return to its source with a failure message. On the other hand, if the docking layer 312 knows that such information is available in its knowledge base 314, it advises the intelligent mobile agent 310a, . . . , 310n, of that fact, and of other relevant capabilities available. For example, in the context of an electromagnetic emitter locator, location 14b might have a database 316 containing raw data representing frequency and pulse characteristics of a received electromagnetic signal, and might also have an expert system 318 which is capable of evaluating the raw data, and which may have already evaluated the raw data, and have identified an emitter as being, for example, a weather radar operating at a particular physical location (which, in general, is not the same as a network location). On the other hand, location 14b might have the raw data represented by database 316, and the knowledge base required as the foundation for an expert system for identifying the emitter, but might not include the expert system itself.
In general, expert systems include two distinct portions, namely a knowledge base and a decision-making portion. The decision-making portion is typically much smaller than the knowledge base. However, the knowledge base changes from time to time as more information becomes available, and as conditions change. For example, in the medical fields the knowledge about the symptoms and progress of newly discovered diseases and/or conditions may not be in an older knowledge base, but, when such information is included in an updated knowledge base, the same types of decision-making programs can operate on the updated information in order to identify, and suggest treatment for, such newly discovered diseases and/or conditions. Similarly, in the abovementioned military application, the knowledge base 318d of FIG. 3 can be updated with the operating frequency, pulse characteristics, and other spectral characteristics of new radar and other emitter types as they become known, which the decision-making portion 318s of the expert system can then use to identify this new type of radar if the raw data is indicative of its presence. Layer portion 320 represents other capabilities which might be available to the intelligent mobile agent.
Thus, the docking layer 312 of FIG. 3 "tells" the then-resident mobile agent or agents of layer 310a, . . . , 310n, about the existence of the raw data available in database 316, and of the existence of either the knowledge base 318d of, or of the entire expert system 318, and also about any other available capabilities 320 which are relevant to the task which the intelligent mobile agent has identified.
Once the intelligent mobile agent 310a, . . . , 310n, of FIG. 3 has received information from the docking layer about the available capabilities, the intelligent mobile agent decides, based upon its program, which of the locally available capabilities it wishes to avail itself of. The docking layer facilitates obtaining the information from the raw-data database or the expert system, or from a stand-alone knowledge base, as required.
The information in expert system database 318 of FIG. 3 can be updated in any of a number of ways. For example, the database 318d can be updated by occasional transmissions over the network 12. While these transmissions may require substantial bandwidth, they are performed infrequently, and may be performed at low data rates, over extended intervals during low-traffic times. In the arrangement of FIG. 3, the expert system and a current database therefor are available to the intelligent mobile agent, without transmitting the entire expert system and its database each time an inquiry is made by transmission of an intelligent mobile agent. The knowledge base 318d may be updated by a memory upload provided by a maintenance worker from a memory disk, or from any other physical media.
FIG. 4 is a simplified flow chart illustrating the operation of a typical mobile agent 310a, . . . , 310n, as described in the abovementioned Whitebread patent application, and FIG. 5 is a simplified flow chart illustrating the operation of the docking system 312. The sequence of events begins with the arrival of a mobile agent at a network node (network site) or an agent dock, as suggested by block 410 of FIG. 4. The sequence representation then flows to a logic node 5-1, which represents a transfer of the sequence to a corresponding logic node 5-1 of FIG. 5. Logic block 510 represents a start point for the docking program, which starts up the logic, and then goes into a "wait" state, waiting for arrival of an agent at the node. From logic node 5-1 of FIG. 5, the logic flows to a block 512, which represents recognition by the logic flow of the docking system 312 that a mobile agent has arrived at the network node. The recognition represented by block 512 may be accompanied by a verification of a security code, following which the docking system logic flows to a decision block illustrated as 514. Decision block 514 decides whether the currently arrived agent has clearance to coact with this network node. If the mobile agent does not meet the security requirement, the logic leaves decision block 514 by the NO output, and flows to a block 518, which represents rejection of the mobile agent. The logic of the agent docking system then returns by a logic path 419 to block 512, whereupon the docking system continues to monitor for the appearance of an active agent at its network node.
If a mobile agent arrives at the network node and meets the established security criteria, the logic leaves decision block 514 of FIG. 5 by way of the YES output, and flows to a block 516, which represents a command for the starting of the mobile agent. From block 516, the logic flows by way of a logic node 4-2 to block 412 of FIG. 4. Block 412 of FIG. 4 represents the starting of the mobile agent in response to the command issued by logic block 516. From logic block 412, the logic flows to a decision block 414, which decides whether the mobile agent requires capability extension. This decision may be as simple as the examination of an "extension required" flag associated with the mobile agent. If no capability extension is required, the logic flows from decision block 414 by way of its NO output, and arrives at a block 416. Block 416 represents performance of the task assigned to the mobile agent by its internal programming. Once the task is completed, the logic flows from block 416 to a block 418. Since the mobile agent has completed its task, it must now decide whether it must return to the originating source with its data, or to seek more data by going on to a known further address, or search randomly by going to an address which has not yet been visited. Block 418 represents the decision as to where the mobile agent is to go next. From block 418, the logic flows to a block 420, which represents an instruction to the docking system to command the transmission of the mobile agent to the selected address. From logic block 420 of FIG. 4, the logic then flows by way of a logic node 5-3 to logic block 520 of FIG. 5, which represents acceptance of the direction to move the mobile agent to a defined network node. Block 522 represents the sending of the mobile agent to its next destination, and the logic then flows by way of a path 523 back to block 512.
In the event that the mobile agent requires extension of its capabilities, the logic flow is as described above until the logic arrives at decision block 414 of FIG. 4. In the case in which the mobile agent requires capability extension, the logic leaves decision block 414 by the YES output, and flows to a block 422, which represents making a request for a particular extension of capabilities from the docking system. It must be recognized that the network may have many nodes having different capabilities, and may have various kinds of mobile agents perambulating therethrough, seeking various different types of capabilities, not all of which are available at each network node. From block 422 of FIG. 4, the logic flows by way of a logic node 5-2 to a block 524 of FIG. 5. Block 524 represents the reception of a request for a particular type of extension capability. From block 524, the logic of the docking system flows to a decision block 526, which looks through its index of available capabilities to see if the requested capability is available. If the capability is not available at this particular network node, the logic leaves decision block 526 by the NO output, and flows to a block 528, which represents the sending of a "not available" signal to the mobile agent. From block 528 of FIG. 5, the control flows by way of a logic node 4-3 to decision block 424 of FIG. 4. Since the capability sought by the mobile agent is not available at this particular network node, the logic of the mobile agent leaves decision block 424 by the NO output, and flows to block 418. As described above, block 418 represents the determination of the location of the next network node to be visited by the mobile agent; from block 418, the logic passes through the states represented by logic blocks or nodes 420, 5-3, 520, and 522, back to block 512, which represents a "waiting for a mobile agent" state of the docking system.
As so far described, the cases which have been discussed in relation to FIGS. 4 and 5 are those in which (a) the mobile agent was not authorized to act, and was rejected; (b) was authorized to act, but did not require capability extension; (c) was authorized to act, and required capability extension, but the particular capability extension was not available at the network node or site. The next possibility is that the mobile agent arriving at the network node is authorized, requires capability, and the capability is available. In this last case, the logic will arrive at decision block 526 of FIG. 5 as described above, but, since the requested capability is available, the logic will leave by the YES output, rather than NO. From the YES output of decision block 526, the logic flows to a block 530, which represents the passing of a capability pointer to the requesting mobile agent. Logic node 4-3 returns the sequence flow to decision block 424 of FIG. 4, whereupon the logic leaves decision block 424 by the YES output, and flows to block 416, which as mentioned above represents the performance of the task, which presumably also means availing itself of the extended capabilities at the network node.
Thus, the arrangement described in the abovementioned Whitebread et al. application allows highly capable intelligent agents of relatively small size to be transmitted over limited-bandwidth networks for performing their functions, with the extended capabilities being made available at the appropriate remote network locations. Portable storage media may be used to take extended capability information to each computer remote computer. In some cases, it may instead be desirable to load the extended capability information into the computer by using the limited-bandwidth network, albeit at a data rate commensurate with network capabilities, by transmission over a period of time, preferably at a time when the network is lightly loaded.
FIG. 6 is a simplified block diagram illustrating three different arrangements for loading extended capability information into a computer at a remote location on the network so that it may be accessed by the docking system of that computer in response to the docking of a mobile agent. In FIG. 6, representative node 14d includes a medium reader, such as a floppy-disk reader 610, which is coupled to a capabilities repository, corresponding to the database, expert system, or the like of FIG. 3. Updates of the capability can be manually transported to the node on a data medium, illustrated as a floppy disk 612, and uploaded by means of the reader 610. Control could be accomplished manually by an operator at a remote site, with the delivery of the update information being performed over the network. It will be clear that updates over the network must be accomplished in the presence of sufficient security to guarantee that unwanted intrusion does not occur. The physical transmission by way of a data medium may also require security measures to assure that the knowledge base is not updated by someone having physical access but without authority to update. Another way to update the knowledge base is by a mobile agent, illustrated as Loader Agent 614, which is especially adapted for uploading or deliver of such information from a remote site during periods of low traffic on the network.
While it is very desirable to be able to express highly capable mobile agents at remote nodes, it is also advantageous to be able to use the cooperation of multiple intelligent mobile agents to perform various tasks. Cooperation of multiple agents implies the transfer of information between the agents. The bandwidth limitation of the communication network prevents the mobile agents from carrying large amounts of information. Improved methods of exchange of information between intelligent mobile agents is desired.